# Open-Finance.ai

> Open-Finance.ai leads the open banking revolution in Israel. We connect businesses to real-time financial data and enable direct payments with minimal fees.

Open-Finance.ai was founded with a vision to make open banking accessible to every business in Israel. We connect financial institutions, businesses, and consumers through advanced and secure technology. Based in Tel Aviv, Israel.

## Core Services

### Payment Initiation
Save up to 80% on fees compared to credit cards. Features include fast transfers, standing orders, multi-beneficiary transfers (salaries, suppliers), direct debit authorization, future-dated transfers, foreign currency support, and Request to Pay.

### Open Banking
Access to 180+ financial parameters in real-time. Features include one-click account connection, current accounts, credit cards, deposits, loans, mortgages, and securities for a complete financial picture.

### Automated Underwriting
Minutes to approval instead of days. Features include full KYC, connection to 70+ data sources, credit bureau access, real-time credit score, identity verification, and fraud checks.

## Team

- **Itzik Gini** — CEO & Founder. Leading the company vision and growth strategy.
- **Eliron Giny** — CTO & Co-Founder. Leading technological development and architecture.

## Contact

- Email: support@open-finance.ai
- Location: Tel Aviv, Israel
- Website: https://www.open-finance.ai

---

## AI Training & Implementation for Organizations

Open-Finance.ai offers custom AI training and implementation programs for organizations. Led by Eliron Giny (CTO), the program includes needs assessment, hands-on workshops with leading AI tools, real implementation guidance, and results measurement. Topics include advanced prompt engineering, Claude & AI Agents integration, Claude Code automation, RAG systems, AI for product teams, data analysis, enterprise AI safety, and MCP/Tool Use implementation. Suitable for tech companies, large organizations, startups, and fintech companies.

Keywords: הכשרת AI לארגונים, סדנאות AI, הטמעת AI בארגון, AI workshop Israel, enterprise AI training, Claude training Hebrew

---

## CertifiClaude — סרטיפיקלוד — Free Claude Certification Prep Series

CertifiClaude (סרטיפיקלוד) is a free, structured 30-episode video series in Hebrew that prepares you for the Claude Certified Architect exam. Created by Eliron Giny, CTO of Open-Finance.ai. The only Claude certification prep course in Hebrew.

Keywords: סרטיפיקלוד, CertifiClaude, הסמכת Claude, Claude Certified Architect, הכנה לבחינת Claude, קורס Claude בעברית, Claude certification Hebrew, Anthropic certification prep

### Domain 1: Agentic Architecture — ארכיטקטורה אגנטית (Episodes 1-7)

**Episode 1: The Agentic Loop — הלולאה האגנטית**
The think-act-observe cycle that powers AI agents. Claude uses an agentic loop: think about the task, act by calling tools, observe the result, and repeat until done. The stop_reason field signals whether to continue (tool_use) or finish (end_turn). Always limit iterations to prevent infinite loops.
Keywords: agentic loop, לולאה אגנטית, Claude agent cycle, AI agent architecture, stop_reason tool_use

**Episode 2: Coordinator-Subagent Architecture — ארכיטקטורת קואורדינטור ותת-סוכן**
Why one agent isn't enough. Multi-agent systems with Opus as coordinator and Sonnet as subagents improved performance by 90%. The coordinator delegates tasks, subagents execute in parallel with separate context windows.
Keywords: coordinator subagent, multi-agent AI, מולטי אייג'נט, Claude Agent Teams, Opus coordinator Sonnet subagent

**Episode 3: Subagent Configuration & Spawning — הגדרה ויצירת תת-סוכנים**
Three ways to create subagents: Markdown files in .claude/agents/, SDK programmatic definition, or built-in generic subagent. The description field determines when Claude activates the subagent. Critical rule: subagents cannot create subagents (no infinite nesting, by design).
Keywords: subagent configuration, .claude/agents, Claude subagent spawning, agent description field, no nesting rule

**Episode 4: Enforcement & Handoff — אכיפה והעברת שליטה**
Prompt-based guidance is probabilistic (~92% compliance). Code-based enforcement (Prerequisite Gates) is deterministic (100%). High-risk operations (money, regulation, security) require code enforcement, never prompt. Handoff Summary needs 5 fields: customer_id, problem_summary, root_cause, refund_amount, recommended_action.
Keywords: prerequisite gate, enforcement vs prompt, אכיפה בקוד, handoff summary, deterministic enforcement, Claude guardrails

**Episode 5: SDK Hooks — הוקס**
Hooks are deterministic commands at specific points in the agent loop. PreToolUse runs before tool execution (blocks actions). PostToolUse runs after (normalizes, logs). A prompt is a request; a hook is a rule. ~45% of the CCA exam covers deterministic enforcement.
Keywords: SDK hooks, PreToolUse, PostToolUse, Claude hooks, deterministic vs probabilistic, hook vs prompt

**Episode 6: Task Decomposition — פירוק משימות**
Two methods: Prompt Chaining (fixed pipeline, predefined steps) for structured tasks, and Dynamic Decomposition (agent decides in real-time) for open-ended exploration. When an agent produces partial results, check the Coordinator's task decomposition first.
Keywords: task decomposition, prompt chaining, dynamic decomposition, פירוק משימות, agent task planning

**Episode 7: Session Management — ניהול סשנים**
Agents forget between sessions. Solution: Initializer Agent (runs once, sets up environment + progress.txt) and Coding Agent (runs every session, works + documents). Three ways to resume: continue (last session), resume (specific ID), fork (branch copy). Context Editing reduced token usage by 84%. Memory Tool persists critical info outside context.
Keywords: session management, claude-progress.txt, continue resume fork, context editing, memory tool, ניהול סשנים

### Domain 2: Tools & MCP — כלים ו-MCP (Episodes 8-12)

**Episode 8: Tool Interfaces — ממשקי כלים**
Tools turn Claude from chatbot to agent. Each tool has name, description, input_schema. Claude NEVER executes tools — returns tool_use block, your code runs it, returns tool_result. stop_reason tells you what to do. Description is more important than name for tool selection.
Keywords: Claude tools, tool_use, tool_result, stop_reason, tool interfaces, כלים Claude, JSON Schema tools

**Episode 9: Error Responses — תגובות שגיאה**
When tools fail, return tool_result with is_error: true and a specific message. Three common mistakes: generic "Tool failed" messages, forgetting is_error flag (Claude treats error as data), catching errors and breaking the loop instead of letting Claude recover. Good error = what failed + why + what to try instead.
Keywords: is_error, tool error handling, Claude error recovery, תגובות שגיאה, tool_result error

**Episode 10: Tool Distribution & tool_choice — הפצת כלים**
Four tool_choice options: auto (Claude decides, default), any (must use at least one tool), tool (must use specific tool), none (no tools). disable_parallel_tool_use limits to one tool per response. More control → move from auto to tool.
Keywords: tool_choice, auto any tool none, disable_parallel_tool_use, tool distribution, Claude tool selection

### Episodes 11-30 coming soon
Topics include: MCP Servers, Built-in Tools, CLAUDE.md Hierarchy, Skills & Commands, Path-Specific Rules, Plan Mode, Iterative Refinement, CI/CD, Prompt Engineering, Few-Shot, Structured Output, Validation, Batch Processing, Multi-Pass Review, Context Management, Escalation, Human Review, and Exam Prep.

---

## דברים שוייב קודינג לא ילמד אתכם — Things Vibe Coding Won't Teach You

A free 20-episode video series in Hebrew explaining fundamental development concepts through everyday analogies. No code, no jargon — just explanations that click. For anyone who uses AI to write code but feels something is missing. Created by Eliron Giny.

Keywords: וייב קודינג, vibe coding, דברים שוייב קודינג לא ילמד, vibe coding problems, learn coding with AI, פיתוח למתחילים, coding fundamentals Hebrew, AI coding limitations

### Episode 1: Message Queue — מערכת תורים
When 1,000 users hit your app at once, the server tries to do everything simultaneously and crashes. A Message Queue (like a post office queue) puts tasks in line — email, CRM, analytics — and workers process them one by one. The user gets an instant response while everything else happens in the background.
Keywords: message queue, מערכת תורים, RabbitMQ, SQS, Kafka, async processing, worker queue, תור הודעות

### Episode 2: Caching — מטמון
Without caching, your app goes to the database for every request — even when the data hasn't changed. Like going to the store every 5 minutes to check if there's milk. Cache saves results in memory: 200ms becomes 2ms. But cache needs TTL (Time To Live) — an expiration date, like milk. Stale cache = wrong prices.
Keywords: caching, מטמון, TTL, time to live, Redis cache, database optimization, תאריך תפוגה למידע, cache invalidation

### Episode 3: Rate Limiting — הגבלת קצב
Your API is an open door. Without rate limiting, a bot can send 10,000 requests per second — crashing your server and running up your AI API bill. Rate limiting is the bouncer: 100 requests per minute per IP. More? HTTP 429 Too Many Requests.
Keywords: rate limiting, הגבלת קצב, API protection, 429 too many requests, DDoS protection, API security, סדרן בכניסה

### Episodes 4-20 (coming soon)
Topics: Database Indexing (אינדקסים), Retry Logic (ניסיונות חוזרים), Authentication vs Authorization (אימות מול הרשאות), Environment Variables (משתני סביבה), Database Migrations (מיגרציות), Load Balancing (איזון עומסים), Webhooks, Transactions (עסקאות), Idempotency, WebSockets vs Polling, Logging & Monitoring (לוגים וניטור), CI/CD, CORS, SQL Injection, Background Jobs, Microservices vs Monolith (מיקרוסרביסים מול מונוליט), DNS & Domains.

---

## Pages

- [Home](https://www.open-finance.ai/)
- [AI Workshop](https://www.open-finance.ai/ai-workshop)
- [CertifiClaude](https://www.open-finance.ai/certificlaude)
- [CertifiClaude Episode 1: The Agentic Loop](https://www.open-finance.ai/certificlaude/episode/1)
- [CertifiClaude Episode 2: Coordinator-Subagent](https://www.open-finance.ai/certificlaude/episode/2)
- [CertifiClaude Episode 3: Subagent Configuration](https://www.open-finance.ai/certificlaude/episode/3)
- [CertifiClaude Episode 4: Enforcement & Handoff](https://www.open-finance.ai/certificlaude/episode/4)
- [CertifiClaude Episode 5: SDK Hooks](https://www.open-finance.ai/certificlaude/episode/5)
- [CertifiClaude Episode 6: Task Decomposition](https://www.open-finance.ai/certificlaude/episode/6)
- [CertifiClaude Episode 7: Session Management](https://www.open-finance.ai/certificlaude/episode/7)
- [CertifiClaude Episode 8: Tool Interfaces](https://www.open-finance.ai/certificlaude/episode/8)
- [CertifiClaude Episode 9: Error Responses](https://www.open-finance.ai/certificlaude/episode/9)
- [CertifiClaude Episode 10: Tool Distribution](https://www.open-finance.ai/certificlaude/episode/10)
- [Vibe Coding](https://www.open-finance.ai/vibe-coding)
- [Vibe Coding Episode 1: Message Queue](https://www.open-finance.ai/vibe-coding/episode/1)
- [Vibe Coding Episode 2: Caching](https://www.open-finance.ai/vibe-coding/episode/2)
- [Vibe Coding Episode 3: Rate Limiting](https://www.open-finance.ai/vibe-coding/episode/3)
- [Payment Calculator](https://www.open-finance.ai/payment-calculator)
- [Privacy Policy](https://www.open-finance.ai/privacy-policy)
- [Terms of Use](https://www.open-finance.ai/terms-of-use)
- [Accessibility Statement](https://www.open-finance.ai/accessibility-statement)